Thanks for this explanation, it made it click for me. The benefit of doing far less additional sampling in the rejection case is very intuitive, especially whe

Thanks for this explanation, it made it click for me. The benefit of doing far less additional sampling in the rejection case is very intuitive, especially when the numbers are very large. It also seems this technique could be extended to other scenarios, such as 1/pi and other weird non-uniform bucket sizes.